Comparison between soft computing methods for tomato quality grading using machine vision

The combination of machine vision and soft computing approaches in the agriculture industry, using training data and automation, can improve processing times by eliminating time consuming manual assessment. The tomato is one of the most popular and highest selling fruits in the world, and its quality is judged by its visual characteristics. Classification of tomatoes into quality grades is therefore very important. In this study, we proposed a series of methods for predicting tomato quality classes based on artificial intelligence. We implemented a multi-layer architecture of a SUB-adaptive neuro fuzzy inference system (MLA-ANFIS) approach using various combinations of multiple input features, neural networks, regression and extreme learning machines (ELMs) based on a tomato image data set with seven input features that were collected from a farm. A deep stacked sparse auto-encoders (DSSAEs) method was proposed for tomato quality grading using image data directly, instead of analysing features extracted from the tomato images. The DSSAEs method was more accurate than previous methods, and used different methodology to previously proposed approaches for the evaluation of the tomato quality grades. The proposed method achieved a sensitivity of 83.2%, specificity of 96.50% and g-mean of 89.40% with accuracy of 95.5%. It may thus be able to improve inspection and quality processing of tomatoes.